A deep learning-based model for drug-target affinity prediction with application to drug repurposing and virtual drug screening

Drug-target affinity prediction is a crucial part of drug discovery. Deep learning-based methods can enable the efficient screening of large drug libraries for candidates which exhibit high binding affinities towards biological targets of interest, thus unveiling drug repurposing opportunities. In this study, a transformer and a convolutional neural network were used to encode molecules and proteins respectively. Subsequently, they were concatenated and fed into a multilayer perceptron in order to create a model which predicts the binding affinity of a molecule to a target of choice. Then, this model was used to screen a drug library for strong binders to SOD1 (superoxide dismutase 1), a protein that has great importance in the disease amyotrophic lateral sclerosis. Among the drugs with the highest predicted binding affinities, there were substances that were bibliographically linked with the target or the disease, making them good candidates for further in vitro and in vivo testing for drug repurposing-related efficacy assessment. The model created has also the potential to streamline the drug screening process, removing the need to screen compounds with low predicted binding affinities as they are unlikely to interact strongly with the target.